Trend Identification using LSTM vs. Convolutional Neural Networks on the NASDAQ market

Capturing broader market trends is essential for making effective and timely trading decisions, yet accurately forecasting these trends remains a significant challenge due to the inherent volatility and dynamic nature of financial data. In this study, we evaluate the performance of Long Short-Term Memory (LSTM) networks in detecting 30-day stock trends using daily closing prices from 100 randomly selected NASDAQ-listed companies. We employ a sliding-window approach to generate input sequences and apply outlier detection to mitigate extreme price fluctuations. Rather than focusing on granular, point-by-point predictions, our emphasis is on identifying overarching trends that drive critical financial decisions. Predictive performance is assessed via the Pearson correlation coefficient, and we compare our LSTM predictions with those obtained from one-dimensional Convolutional Neural Networks (1D CNNs) under identical experimental conditions. Our findings demonstrate that LSTMs consistently deliver more robust trend identification performance than CNNs, highlighting their potential for capturing general market behavior.